Molecular basis for mid-region amyloid-β capture by leading Alzheimer's disease immunotherapies

Solanezumab (Eli Lilly) and crenezumab (Genentech) are the leading clinical antibodies targeting Amyloid-β (Aβ) to be tested in multiple Phase III clinical trials for the prevention of Alzheimer's disease in at-risk individuals. Aβ capture by these clinical antibodies is explained here with the first reported mid-region Aβ-anti-Aβ complex crystal structure. Solanezumab accommodates a large Aβ epitope (960 Å2 buried interface over residues 16 to 26) that forms extensive contacts and hydrogen bonds to the antibody, largely via main-chain Aβ atoms and a deeply buried Phe19-Phe20 dipeptide core. The conformation of Aβ captured is an intermediate between observed sheet and helical forms with intramolecular hydrogen bonds stabilising residues 20–26 in a helical conformation. Remarkably, Aβ-binding residues are almost perfectly conserved in crenezumab. The structure explains the observed shared cross reactivity of solanezumab and crenezumab with proteins abundant in plasma that exhibit this Phe-Phe dipeptide.

Ab residues 16-18 are in an extended coil conformation laying flat over the solanezumab surface, whilst residues C-terminal to the Phe19-Phe20 core, project out of the antibody in a helical conformation from residue Ala21 to Ser26 (Fig. 1). This helix is stabilised by putative hydrogen bonds between Phe20(CO) and Asp23(NH), Ala21 (CO) and Val24 (NH), Asp23(CO) and Ser26(NH). The Phe20-Asp23 H-bond holds the turn posing the helical C-terminal region at a right angle to the coil N-terminal region. There are two putative polar contacts stabilising the Ab conformation N-terminal to the Phe19-Phe20 dipeptide: between Leu17(CO) and the side-chain amine of Lys16 and the main-chain amine of Phe19.

Discussion
This structure is unique amongst published anti-Ab structures. A slew of anti-N-terminal antibody structures holding Ab in an extended coil over the first eight or so residues have been reported [15][16][17][18][19] . We, and others, reported the bapineuzumab and its murine parent 3D6 structures, showing the N-terminal five residues of Ab captured by these antibodies in a helical conformation with a buried N-terminus [20][21][22] . The ponezumab (Pfizer) structure, a failed clinical antibody with specificity for the C-terminus of Ab40, was shown to grasp the highly hydrophobic region (30-AIIGLMVGGVV-40) in an extended coil conformation 23 .   The structure shown here is the first anti-Ab antibody structure targeting the central, oligomer-nucleation core. Much of what we know of the structural biology of this highly pleomorphic peptide has been deduced from NMR studies where solution conditions are artificially manipulated with non-polar solvents such as hexafluoroisopropanol (HFIP) and detergents such as SDS to mimic membrane environments and to shift helical content of the peptide's structure 24,25 . We also have crystallographic models of Ab peptides packed into sheet structures as proposed for oligomeric assemblies and fibrils 26,27 . The structure reported here represents an intermediate structure between helical and sheet forms. Figure 3 shows the Ab structure recognised by solanezumab and reveals that over the KLVF region the peptide adopts a conformation compatible with crystallographic b-sheet models of oligomerisation. This oligomerisation motif is disrupted by a 180u rotation in the psi torsion angle of Phe19, initiating the helical conformation consistent with NMR-derived Ab solution structures, determined in solvents mimicking membrane environments. This helical conformation is adopted by residues Phe20 to Ser26, stabilised by intramolecular, residue i to i13, hydrogen-bonds. Solanezumab has been shown to inhibit fibril formation by synthetic Ab 4 , and only recognises soluble monomeric Ab 28 , which is consistent with the idea that this central epitope helical structure, if present in solution, would be a natural potential energy barrier to oligomerisation and involved early in the process of Ab oligomerisation, becoming unavailable to solanezumab either by the epitope being buried or because of conformational change.
The antigen buried surface area (BSA 29 ) of the Ab epitope recognised by solanezumab is 960 Å 2 , much larger than for Ab epitopes engaged by other antibodies. For example, the N-terminal-directed antibody WO2:Ab complex (and the homologues with protein data bank (PDB) identifiers PFA1, PFA2, 12A11, 10D5, and 12B4 Ab complexes) 15,16,18 showed an epitope with a BSA of ,727 Å 2 . The bapineuzumab structure shows that it captures the N-terminus of Ab, burying the first five residues in a helical conformation 20-22 with a BSA of 537 Å 2 . Gantenerumab recognises a larger N-terminal epitope across Ab residues 1-11 in an extended coil, but its interface area cannot be evaluated as the model is not publically available. Gantenerumab reportedly binds different aggregation states of Ab from 0.6 nM affinity for monomers, to 17 nM affinity for fibrils 17 . The interface ponezumab (PDB id: 3U0T) makes with the hydrophobic C-terminus (residues 30-40) of Ab is 631 Å 2 and that antibody has a 0.3 nM affinity for wild type Ab (residues 1-40) 23 . Typically antigen BSA's for antibody:peptide complexes fall between ,400 Å 2 and 700 Å 2 and hence solanezumab's engagement of Ab is atypical for antibody recognition of peptides 30 . The extensive contacts, including polar contacts, made by solanezumab over a large surface area of Ab is consistent with solanezumab's very high (picomolar) affinity for its ligand 13 . One notable feature of the solanezumab structure is that it has the minimum length for the hypervariable H3 loop in the complementarity determining region (CDR) with just four amino acids in that loop. This truncated H3 loop opens up the ligand binding site, enabling extended engagement of Ab towards its N-terminal end.
One compelling feature of the complex structure is that for the first time we can compare, in detail, Ab engagement by solanezumab and crenezumab ( Fig. 2 and Supp. Fig. 2). We have previously noted that the CDRs of solanezumab and crenezumab are highly homologous 13 is shown bound to solanezumab through its CDRs. The CDRs are represented as a surface with Ab-contacting residues coloured blue. Polar contacts are exhibited as yellow dashed lines. The CDR sequences (L1, L2 and L3 from the light chain, and H1, H2 and H3 from the heavy chain) of solanezumab (sola.) and the clinical immunotherapy crenezumab (crene.) are shown at the bottom of the figure. Each CDR loop in solanezumab is the same length as its counterpart in crenezumab. Antibody residues that contact Ab in the solanezumab-Ab complex structure, and the corresponding residues in crenezumab, are coloured blue. The crenezumab-Ab complex structure was derived by homology modelling from the solanezumab-Ab complex crystal structure (see text). Only two of those contacting residues are not conserved: namely, Sola. residues Phe36 (L1) and Ser33(H1). These are labelled in (a) and their local environments are highlighted respectively in (b) and (c).   in terms of sequence identity, despite having purportedly different relative affinity for monomeric, oligomeric and fibrillar forms. While solanezumab (and parent 266) are known to bind monomeric soluble Ab only, crenezumab has been described as having high affinity for monomeric, oligomeric and fibrillar forms 3 . All CDRs are identical in length to their counterpart in solanezumab and crenezumab (Fig. 2). Three are identical in composition; namely, L2, L3 and H3, and each of these make significant contact with Ab. The least conserved CDR (H2) does not at all contact Ab. The remaining L1 and H1 CDRs each have one non-conservative mutation in Ab-contacting residues. The only two interesting differences between solanezumab and crenezumab, besides their isotypes (IgG1 vs IgG4), are at Ser33(H1) and Phe36(L1) (technically just outside L1), which are tyrosine and glycine, respectively, in crenezumab. Mutagenesis/affinity measurements are required to confirm the relative importance of the two residues, but the former would result in a loss of one of three Ab side-chain H-bonds made with solanezumab and the latter introduces a polar hydroxyl moiety into the core hydrophobic cavity engaging the side-chains of Phe19 and Phe20. These differences can account in large part for the significant difference in affinity of Ab reported for crenezumab (low nM) and solanezumab (pM) 13 . However, modelling suggests these changes are unlikely to significantly impact the conformation of the large Ab epitope recognised by these antibodies.
The final aspect that this structure explains is the basis of cross reactivity of these antibodies with other proteins as was recently reported by us 13 . IP pull downs and MS/MS studies led to the identification of a dozen proteins in AD-affected plasma recognised by both solanezumab and crenezumab, with magnetic beads alone and bapineuzumab coated beads as negative controls. The plasma proteins identified as cross reacting with solanezumab and crenezumab share identity with the Ab KLVFF epitope, which is the core of the Ab epitope observed in our structure (Fig. 1). Given that much of the engagement by solanezumab of Ab is via the side-chains of some of these core residues plus extensive interactions with the larger peptide via main-chain elements, it is not surprising that there are cross reactivity issues with more abundant proteins in AD-affected tissue displaying substantial parts of this linear epitope.
The structure described here provides a basis for the design of next generation antibodies with diminished cross reactivity potential. Importantly, the identification of alternative mechanisms of action of solanezumab, and crenezumab, through engagement with proteins sharing elements of the Ab epitope, gives insights into alternative therapeutic pathways for AD, if reported cognitive benefit in the absence of amyloid reduction with solanezumab treatment is reproduced in upcoming AD prevention trials.

Methods
Protein production. DNA corresponding to the Fab portion of solanezumab (defined in (Ref. 31) and elsewhere: Patent WO 2001062801 A2, CAS 955085-14-0, CHEMBL1743072) with a C-terminal hexa-histidine tag on the heavy chain was synthesised (Genscript). These DNA constructs were cloned into pcDNA3.11 expression plasmids. Single point mutation was performed to replace the glycosylation site in Asn55(H2)Ser to facilitate crystallisation. Heavy and light chain constructs were co-transfected into FreeStyle TM 293-F cells (Invitrogen). Cell culture supernatants were harvested by centrifugation and concentrated by tangential flow filtration (Millipore, Proflux M12). Fab was purified with Ni-NTA resin (Qiagen) followed by size exclusion chromatography, dialysed extensively against Buffer A (20 mM HEPES pH 7.5 and 50 mM NaCl), and finally concentrated to 5 mg/mL and stored in small aliquots at -80uC until required for crystallisation.
Fab-Ab complex preparation. Peptide corresponding to residues 12-28 (Ab 12-28 ) of the wild type amyloid-b sequence (DAEFRHDSGYE-12 VHHQKLVFFAEDVGSNK 28 -GAIIGLMVGGVV) was purchased from Anaspec (95% purity). Peptide was resuspended in Milli-Q water and aliquoted to give 100 mg per Eppendorf tube. Peptide was added to antibody to a Fab:Ab molar ratio of 152 and dialysed in 10 mM HEPES pH 7.5 for 4 hours.
Crystallisation. Crystallisation trials of Fab:Ab12-28 complex was set up manually using a low ionic strength screen 32 and the hanging-drop vapor-diffusion method in 24-well greased plates (Hampton Research) at 295 K. In each crystallisation drop, 1 mL of PEG 3350 (from 4 to 24% w/v) and 1 mL of 50 mM low ionic buffer were added to 2 mL protein solution. The protein droplets were equilibrated with 500 mL of ,24% w/v PEG3350 reservoir solution to ensure a fast evaporation rate. The best crystals obtained were grown in 16% w/v PEG 3350 and 50 mM sodium citrate pH 4. Crystals were harvested after 2 weeks and then soaked for 30 seconds in a cryoprotectant (25% v/v of glycerol and drop solution), cryocooled in liquid nitrogen, and mounted in a cryostream at 100 K for data collection.
Data collection and structural determination. X-ray diffraction data were acquired at the MX2 beamline at the Australian Synchrotron (Clayton, Victoria). Data collection was controlled using Blu-Ice software 33 . A data set of 720 images was acquired at a wavelength of 0.9537 Å , with 0.5u rotation per frame. The data set was processed with XDS 34 and scaled in point group P1 using Aimless of the CCP4 suite 35 . Five per cent of the reflections were set aside by Aimless for the free R set.
The initial structure was determined by molecular replacement with Phaser and Molrep of the CCP4 suite 35 . A successful molecular replacement solution was achieved with a probe model derived from the crystal structure of a humanised 3D6 Fab bound to amyloid beta peptide, PDB entry code 4HIX 20 , identified in a Protein Data Bank search based sequence similarities to humanised solanezumab. The successful search identified two copies of the complex in the asymmetric unit. Several rounds of refinement were done with Buster (Global Phasing Ltd) including TLS and individual isotropic B-factor refinement. TLS refinement was necessary as the data were anisotropic to 2.8 Å in the a* direction, but remained at 2.4 Å in the remaining two directions. Rebuilding was performed using Coot 36 . Water molecules were added if they had good spherical density, favourable hydrogen bonding, and reasonable Bfactors. Well-defined density for Ab peptide residues 16 to 26 and 16 to 24 were immediately identified in each molecule of the asymmetric unit respectively; however, this was not modelled until the protein structure was nearing completion. Structure validation was monitored with MolProbity 37 . An homology model of crenezumab was constructing using the solanezumab crystal structure as a guide.